QQ客服 官方微博 反馈留言

	Accelerating GPGPU architecture simulation
	Yu zhibin; Eeckhout Lieven; Goswami Nilanjian; Li tao; John lizy K.; Jin hai; Xu chengzhong
	2013
会议名称	2013 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, SIGMETRICS 2013
会议地点	Pittsburgh, PA
英文摘要	Recently, graphics processing units (GPUs) have opened up new opportunities for speeding up general-purpose parallel applications due to their massive computational power and up to hundreds of thousands of threads enabled by programming models such as CUDA. However, due to the serial nature of existing micro-architecture simulators, these massively parallel architectures and workloads need to be simulated sequentially. As a result, simulating GPGPUarchitectures with typical benchmarks and input data sets is extremely time-consuming. This paper addresses the GPGPU architecture simulation challenge by generating miniature, yet representative GPGPU kernels. We first summarize the static characteristics of an existing GPGPU kernel in a profile, and analyze its dynamic behavior using the novel concept of the divergence flow statistics graph (DFSG). We subsequently use a GPGPU kernel synthesizing framework to generate a miniature proxy of the original kernel, which can reduce simulation time significantly. The key idea is to reduce the number of simulated instructions by decreasing per-thread iteration counts of loops. Our experimental results show that our approach can accelerate GPGPU architecture simulation by a factor of 88X on average and up to 589X with an average IPC relative error of 5.6%.(4 refs)
收录类别	EI
语种	英语
内容类型	会议论文
源URL	[http://ir.siat.ac.cn:8080/handle/172644/5124]
专题	深圳先进技术研究院_数字所
作者单位	2013
推荐引用方式 GB/T 7714	Yu zhibin,Eeckhout Lieven,Goswami Nilanjian,et al. Accelerating GPGPU architecture simulation[C]. 见:2013 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, SIGMETRICS 2013. Pittsburgh, PA.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。